This patch allows the LPFC to start up without a fatal kernel bug based
on an exceeded KMALLOC_MAX_SIZE and a too large NR_CPU-based maskbits
field. The bug was based on the number of CPU cores in a system.
Using the get_cpu_mask() function declared in kernel/cpu.c allows the
driver to load on the community kernel 4.2 RC1.
Below is the kernel bug reproduced:
8<--------------------------------------------------------------------
2199382.828437 ( 0.005216)| lpfc 0003:02:00.0: enabling device (0140 -> 0142)
2199382.999272 ( 0.170835)| ------------[ cut here ]------------
2199382.999337 ( 0.000065)| WARNING: CPU: 84 PID: 404 at mm/slab_common.c:653 kmalloc_slab+0x2f/0x89()
2199383.004534 ( 0.005197)| Modules linked in: lpfc(+) usbcore(+) mptctl scsi_transport_fc sg lpc_ich i2c_i801 usb_common tpm_tis mfd_core tpm acpi_cpufreq button scsi_dh_alua scsi_dh_rdacusbcore: registered new device driver usb
2199383.020568 ( 0.016034)|
2199383.020581 ( 0.000013)| scsi_dh_hp_sw scsi_dh_emc scsi_dh gru thermal sata_nv processor piix fan thermal_sysehci_hcd: USB 2.0 'Enhanced' Host Controller (EHCI) Driver
2199383.035288 ( 0.014707)|
2199383.035306 ( 0.000018)| hwmon ata_piix
2199383.035336 ( 0.000030)| CPU: 84 PID: 404 Comm: kworker/84:0 Not tainted
3.18.0-rc2-gat-00106-ga7ca10f-dirty #178
2199383.047077 ( 0.011741)| ehci-pci: EHCI PCI platform driver
2199383.047134 ( 0.000057)| Hardware name: SGI UV2000/ROMLEY, BIOS SGI UV 2000/3000 series BIOS 01/15/2013
2199383.056245 ( 0.009111)| Workqueue: events work_for_cpu_fn
2199383.066174 ( 0.009929)|
000000000000028d ffff88eef827bbe8 ffffffff815a542f 000000000000028d
2199383.069545 ( 0.003371)|
ffffffff810ea142 ffff88eef827bc28 ffffffff8104365c ffff88eefe4006c8
2199383.076214 ( 0.006669)|
0000000000000000 00000000000080d0 0000000000000000 0000000000000004
2199383.079213 ( 0.002999)| Call Trace:
2199383.084084 ( 0.004871)| [<
ffffffff815a542f>] dump_stack+0x49/0x62
2199383.087283 ( 0.003199)| [<
ffffffff810ea142>] ? kmalloc_slab+0x2f/0x89
2199383.091415 ( 0.004132)| [<
ffffffff8104365c>] warn_slowpath_common+0x77/0x92
2199383.095197 ( 0.003782)| [<
ffffffff8104368c>] warn_slowpath_null+0x15/0x17
2199383.103336 ( 0.008139)| [<
ffffffff810ea142>] kmalloc_slab+0x2f/0x89
2199383.107082 ( 0.003746)| [<
ffffffff8110fd9e>] __kmalloc+0x13/0x16a
2199383.112531 ( 0.005449)| [<
ffffffffa01a8ed9>] lpfc_pci_probe_one_s4+0x105b/0x1644 [lpfc]
2199383.115316 ( 0.002785)| [<
ffffffff81302b92>] ? pci_bus_read_config_dword+0x75/0x87
2199383.123431 ( 0.008115)| [<
ffffffffa01a951f>] lpfc_pci_probe_one+0x5d/0xcb5 [lpfc]
2199383.127364 ( 0.003933)| [<
ffffffff81497119>] ? dbs_check_cpu+0x168/0x177
2199383.136438 ( 0.009074)| [<
ffffffff81496fa5>] ? gov_queue_work+0xb4/0xc0
2199383.140407 ( 0.003969)| [<
ffffffff8130b2a1>] local_pci_probe+0x1e/0x52
2199383.143105 ( 0.002698)| [<
ffffffff81052c47>] work_for_cpu_fn+0x13/0x1b
2199383.147315 ( 0.004210)| [<
ffffffff81054965>] process_one_work+0x222/0x35e
2199383.151379 ( 0.004064)| [<
ffffffff81054e76>] worker_thread+0x3d5/0x46e
2199383.159402 ( 0.008023)| [<
ffffffff81054aa1>] ? process_one_work+0x35e/0x35e
2199383.163097 ( 0.003695)| [<
ffffffff810599c6>] kthread+0xc8/0xd2
2199383.167476 ( 0.004379)| [<
ffffffff810598fe>] ? kthread_freezable_should_stop+0x5b/0x5b
2199383.176434 ( 0.008958)| [<
ffffffff815a8cac>] ret_from_fork+0x7c/0xb0
2199383.180086 ( 0.003652)| [<
ffffffff810598fe>] ? kthread_freezable_should_stop+0x5b/0x5b
2199383.192333 ( 0.012247)| ehci-pci 0000:00:1a.0: EHCI Host Controller
-------------------------------------------------------------------->8
The proposed solution was approved by James Smart at Emulex and tested
on a UV2 machine with 6144 cores. With the fix, the LPFC module loads
with no unwanted effects on the system.
Signed-off-by: Ian Mitchell <imitchell@sgi.com>
Signed-off-by: Alex Thorlton <athorlton@sgi.com>
Suggested-by: Robert Elliot <elliott@hp.com>
[james.smart: resolve unused variable warning]
Signed-off-by: James Smart <james.smart@avagotech.com>
Signed-off-by: James Bottomley <JBottomley@Odin.com>
#ifdef CONFIG_X86
struct cpuinfo_x86 *cpuinfo;
#endif
- struct cpumask *mask;
uint8_t chann[LPFC_FCP_IO_CHAN_MAX+1];
/* If there is no mapping, just return */
first_cpu = cpu;
/* Now affinitize to the selected CPU */
- mask = &cpup->maskbits;
- cpumask_clear(mask);
- cpumask_set_cpu(cpu, mask);
i = irq_set_affinity_hint(phba->sli4_hba.msix_entries[idx].
- vector, mask);
+ vector, get_cpu_mask(cpu));
lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
"3330 Set Affinity: CPU %d channel %d "
uint16_t core_id;
uint16_t irq;
uint16_t channel_id;
- struct cpumask maskbits;
};
#define LPFC_VECTOR_MAP_EMPTY 0xffff